关键词: alpha稳定分布, 独立分量分析, 盲信号分离, 重拖尾信号, 多层神经网络

Abstract: The convention Blind Source Separation（BSS） algorithms have a poor separating effect for the heavy-tailed signals. An online BSS algorithm for heavy-tailed signals is proposed based on the alpha stable distribution theory. The algorithm estimates the Probability Density Functions（PDFs） of the standard alpha stable distribution, and establishes the standard function library look-up table. It estimates on-line the signal characteristic parameter, symmetrical parameter and scale parameter, and therefore the PDF is obtained fast. The score function is obtained accurately by using the multilayer neural networks. The simulated results show that, the proposed algorithm has a well convergence property and a lower computational complexity.

Key words: alpha stable distribution, independent component analysis, blind source separation, heavy-tailed signals, multilayer neural networks